Alibaba Teases Qwen4 Architecture via Qwen3.8-Flash-Next

Alibaba Cloud previewed architectural innovations for its upcoming Qwen4 foundation model family through the release of Qwen3.8-Flash-Next. The architecture features extreme sparse Mixture-of-Experts routing designed to deliver the knowledge capacity of massive models with the inference latency of lightweight models. The weights incorporate native agent tool-calling hooks, optimizing performance on multi-step reasoning and automated API execution.